Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchAI, startups & growth insights. No spam.
TL;DR: California legislators have introduced a bill that would end the state's 50-year nuclear construction moratorium — driven almost entirely by the power demands of AI data centers. U.S. data center electricity demand is expected to nearly double from 19 GW in 2023 to 35 GW by 2030. Meta, Microsoft, Google, and Amazon have all signed major nuclear deals in the past 18 months. California, home to the largest concentration of AI companies in the world, can no longer afford to be the state that says no to the only carbon-free energy source that runs around the clock.
In early March 2026, California state legislators introduced a bill that, if passed, would effectively nullify the nuclear construction moratorium that has been on the books since the late 1970s. The legislation would allow next-generation nuclear technologies — including small modular reactors, advanced fission designs, and other configurations that have received federal licensing since 2005 — to be sited, permitted, and built inside state lines for the first time in nearly five decades.
This is not a minor regulatory tweak. It is the unwinding of one of the most entrenched energy policy positions of any state in the country. California's moratorium was not accidental. It was a deliberate, politically charged decision that reflected the dominant environmental consensus of its era. Reversing it requires overcoming that history, the constituencies that still defend it, and a regulatory infrastructure that has not processed a nuclear application in two generations.
What has changed is not the technology or the politics in any traditional sense. What has changed is the scale of the electricity demand that AI infrastructure is about to impose on the California grid, and the realization that there is no credible pathway to meet that demand without some form of always-on, zero-carbon generation — which in practice means nuclear.
The Bloomberg report published March 2, 2026 described the legislation in the context of a broader national conversation about energy supply and AI workloads. California is not alone in reconsidering its nuclear posture. Virginia, Georgia, and Tennessee are all actively discussing nuclear expansion for the same reason. But California moving is the headline because California has the longest, loudest history of opposition — and because California is home to Anthropic, Google DeepMind, OpenAI, NVIDIA, and the largest concentration of AI infrastructure investment in the world.
The moratorium that may now end was born in a different world. Ending it is a statement that the AI era has arrived, and it is hungry.
To understand what California is reversing, you have to understand what California was reacting to in the 1970s.
The state's nuclear moratorium was not passed in a vacuum. It came at the tail end of a decade in which nuclear energy had gone from being the clean, cheap, futuristic energy source of the postwar consensus to a source of deep public anxiety. The 1970s oil crisis had reshuffled energy politics globally. Anti-nuclear activism had become a mass movement. And the regulatory environment was shifting in ways that made nuclear projects increasingly expensive and legally uncertain.
The decisive moment came on March 28, 1979: the Three Mile Island accident in Pennsylvania. A partial meltdown at the Unit 2 reactor released a small amount of radioactive gas and triggered one of the largest civil evacuations in American history. No one died directly from the accident, but its psychological and political effect was permanent. Nuclear energy ceased to be bipartisan. In California, where the anti-nuclear movement was already strong, Three Mile Island provided the political catalyst for what became a formal moratorium on new construction.
The moratorium did not shut down existing plants. It blocked new ones. California's position was that no new nuclear generation could be added until the federal government demonstrated a solution to permanent radioactive waste storage — a standard that, as of the 1970s, no one could meet and that remained unmet for decades.
That framing was important. The moratorium was not technically a permanent ban. It was conditional on waste storage progress. But in practice, it functioned as a permanent ban because the condition was never met. The federal government's nuclear waste program stalled at Yucca Mountain and has not recovered. California had effective cover to maintain its position without ever having to formally defend it.
For 50 years, that worked. Renewables, natural gas, and imported power allowed California to grow without building new nuclear. The state positioned itself as the national leader in solar and wind development. The moratorium became part of California's environmental identity.
Then the AI data centers arrived.
The Federal Energy Regulatory Commission (FERC) has published projections that describe what AI infrastructure is going to do to the U.S. electricity grid. The numbers are not ambiguous.
U.S. data center electricity demand stood at approximately 19 gigawatts in 2023. FERC projects that demand to reach 35 GW by 2030 — nearly doubling in seven years. For context, 35 GW is roughly equivalent to the total generating capacity of a medium-sized European country. The United States is about to add that much dedicated electricity demand just for computing infrastructure.
| Year | U.S. Data Center Demand | Key Driver |
|---|---|---|
| 2023 | 19 GW | Cloud + pre-AI workloads |
| 2025 | ~25 GW | LLM training + inference scaling |
| 2027 | ~29 GW | Agentic AI, multimodal deployments |
| 2030 | 35 GW (projected) | Frontier model inference at scale |
The AI component of this demand is not evenly distributed. Training a single frontier model — a GPT-4 class system or larger — consumes electricity at a rate comparable to tens of thousands of homes. Inference is more distributed but also far more continuous. Unlike a household, a data center running AI inference at scale runs at near-100% capacity utilization 24 hours a day, 365 days a year. It does not sleep. It does not take weekends off. It does not have seasonal demand variation. It just runs.
That load profile is almost uniquely difficult for renewable energy to serve without storage. Solar generates power during daylight hours. Wind is intermittent by definition. Both require either massive battery storage or gas peaker plants to fill the gaps. Battery storage at grid scale is improving rapidly but remains expensive and limited in duration — current utility-scale installations typically provide four to eight hours of storage, which is not enough to cover extended periods of low renewable generation.
California's grid, already under stress during summer heat waves, is being asked to absorb a second growth curve on top of normal demand growth — one that does not respond to time-of-use pricing or demand response incentives because AI inference cannot simply be paused for economic reasons when a critical model is serving live users.
The math on reaching 35 GW of clean, reliable power without nuclear does not close easily, especially in a state that has already deployed most of its accessible renewable sites.
The energy policy conversation around AI has a notable pattern: every proposed solution gets debuted, analyzed, and then quietly acknowledged to be insufficient as a standalone answer.
Solar at scale runs into land use constraints, permitting timelines, and the night problem. Wind faces similar intermittency issues plus the geographic mismatch between good wind resources and data center locations. Battery storage is advancing quickly but is not yet able to provide multi-day backup at the scale AI data centers require. Geothermal is promising but geographically limited. Hydrogen is early-stage and expensive. Efficiency improvements matter but cannot bend the curve enough to absorb the AI load.
Nuclear does not have any of these constraints. A nuclear plant runs at over 90% capacity factor — meaning it produces power more than 90% of the hours in a year. It does so regardless of weather. It produces no direct carbon emissions. Its fuel is energy-dense enough that a year's worth of fuel for a reactor fits in a space that would hold a few hundred barrels of oil. A single large reactor produces 1 GW of continuous power — the equivalent of hundreds of thousands of solar panels or wind turbines operating at average capacity.
The counterarguments to nuclear are real: construction costs are high, timelines are long, waste storage remains unresolved, and the permitting process has historically been brutal. But the counterarguments to every alternative in the context of AI power demand are also real, and they are more fundamental. You can build nuclear on any site with adequate water access. You cannot build sufficient solar in a data-center-dense corridor where land costs are prohibitive. You cannot build enough wind in northern California to power San Jose.
For AI specifically, the capacity factor argument is decisive. Data centers need power that is available every hour of every year, not just when the sun is shining. Nuclear is the only zero-carbon technology that delivers that profile at scale with current technology. The AI industry has collectively reached this conclusion, as evidenced by what it has been doing with its checkbook.
The clearest signal that the technology industry has concluded nuclear is the answer is where it is signing contracts.
Meta signed a 20-year nuclear power purchase agreement with Constellation Energy, tying its data center infrastructure to the Clinton Clean Energy Center in Illinois. A 20-year deal is not a hedging position. It is a statement of conviction that nuclear will be the foundation of AI infrastructure power for the next generation.
Microsoft went further in symbolism if not scale: it struck a deal to restart Three Mile Island — the very plant whose 1979 accident triggered California's moratorium. The restarted unit, now called the Crane Clean Energy Center, is supplying power to Microsoft's data centers running Copilot and Azure AI workloads. The optics of using Three Mile Island to power AI are not accidental. Microsoft is making a statement about how much the risk calculus around nuclear has changed.
Google signed what was announced as the first-ever commercial small modular reactor agreement, partnering with Kairos Power to develop SMR capacity for Google Cloud infrastructure. Google's position is notable because Google is also California's largest corporate tenant and one of the state's biggest electricity consumers. If Google is betting on SMRs nationally, it has obvious interest in California's moratorium falling.
Amazon has pursued nuclear power agreements for AWS data centers through multiple channels, including agreements with Energy Northwest for SMR development and investments in nuclear startups. AWS is the largest cloud provider globally; its power needs are proportionally enormous.
| Company | Nuclear Deal | Scale |
|---|---|---|
| Meta | Constellation Energy (Clinton, IL) | 20-year PPA |
| Microsoft | Three Mile Island restart | ~835 MW |
| Kairos Power SMR agreement | Multiple units | |
| Amazon | Energy Northwest + startup investments | SMR pipeline |
These are not pilot projects or PR gestures. They are infrastructure commitments at the scale that tech companies make when they believe a technology is going to be load-bearing for their business for decades.
The technologies that California's moratorium most directly blocked are not the large light-water reactors of the 1970s that triggered the original opposition. They are small modular reactors — a category of advanced nuclear designs that received their first federal licensing approvals starting in 2005, more than 25 years after California's moratorium took effect.
SMRs are fundamentally different from the reactors the public imagines when they hear "nuclear power plant." They are smaller, by design — typically under 300 MW per unit, compared to the 1,000+ MW of a conventional large reactor. They can be factory-manufactured and shipped to site, rather than built piece by piece over years of on-site construction. Their safety systems are designed to be passive — meaning they use physics rather than pumps and operators to maintain safe temperatures in an emergency, eliminating the human error scenarios that contributed to Three Mile Island and Chernobyl.
Some SMR designs do not use water cooling at all, removing the need for coastal or riverside siting. Some use fuel cycles that produce substantially less long-lived waste. A few designs can consume existing nuclear waste as fuel, potentially addressing the waste storage issue that California explicitly cited as its condition for lifting the moratorium.
The irony is significant. California's stated reason for the moratorium — the absence of a federal waste solution — is closer to being addressed by new reactor designs than by anything the federal government has built at Yucca Mountain. And the technologies are already federally licensed. The barrier is purely California's own law.
The new legislation would allow these federally approved technologies to move through California's permitting process for the first time. That is not the same as building a reactor tomorrow. California's permitting infrastructure has not reviewed a nuclear application in 50 years. But it opens the door.
Nuclear energy has traditionally been a political fault line: Republicans for it, Democrats against it, especially in California. The new legislation suggests that map is changing, driven by economics that do not respect partisan alignment.
The AI industry is overwhelmingly concentrated in Democratic-voting, environmentally conscious California. The founders and executives of Anthropic, OpenAI, Google, and Meta are not culturally aligned with the traditional pro-nuclear Republican coalition. But they are running companies that are about to consume as much electricity as small countries, and they live in a state that cannot build enough renewable capacity to serve that demand reliably.
That creates a political constituency for nuclear that did not exist in 2010: wealthy, politically connected tech executives who need reliable clean power and are willing to advocate for it, along with the labor unions that represent construction workers who would build nuclear plants and the utilities that see nuclear as a revenue opportunity.
The opposition coalition is also shifting. Traditional environmental groups opposed to nuclear are now facing the argument that the alternative to nuclear for AI power demand is natural gas — a position that is harder to defend on climate grounds than it once was. Some major environmental organizations, including voices within the Sierra Club, have moved toward "conditional support" for advanced nuclear as a climate necessity.
Virginia, Georgia, and Tennessee are all considering nuclear expansion explicitly linked to data center demand. In each case, the political dynamic is the same: tech company power demand is overriding the previous political calculus on nuclear.
The environmental opposition to California's nuclear reversal is real and organized, but it is arguing from a weakened position.
The traditional anti-nuclear arguments — accident risk, waste storage, cost overruns, proliferation — remain factually grounded. Nuclear accidents, while rare, have lasting consequences. The U.S. has no operational permanent repository for high-level nuclear waste. Large nuclear construction projects have a history of dramatic cost and schedule overruns, as the Vogtle expansion in Georgia demonstrated painfully. Enriched uranium presents proliferation risks that coal and solar panels do not.
But the environmental groups making these arguments are now in a conversation where the opposing side is not utilities or fossil fuel companies. It is the companies building the technology that is driving the energy demand — companies whose products many environmentalists use and support, backed by people who have made major donations to environmental causes, arguing that nuclear is necessary specifically to avoid burning more gas.
The debate has become genuinely uncomfortable for the environmental movement. If you oppose nuclear in California and the result is that AI data centers get powered by natural gas peakers, you have made a climate trade-off that is hard to defend on the numbers. If you support nuclear, you are reversing a half-century of movement orthodoxy.
Some groups are threading this needle by supporting SMR research while opposing near-term deployment — a position that satisfies neither side of the debate and is becoming harder to sustain as the gap between AI power demand and available clean supply grows visibly.
The introduction of the legislation is the beginning of a process that will take years even under the most optimistic scenario.
California's legislative process requires bills to pass committee hearings, floor votes in both chambers, and signature by the governor before becoming law. For legislation this politically significant, that process will involve extensive public comment, environmental review under CEQA (California Environmental Quality Act), and likely legal challenges from opposition groups.
Even if the bill passes and is signed in 2026, lifting the moratorium does not immediately produce electricity. A utility or developer would still need to select a site, complete environmental review, obtain federal licensing from the NRC (which for SMR designs that are already licensed is faster but not instantaneous), and build the plant. SMR construction timelines are estimated at 3–5 years per unit for factory-manufactured designs under ideal conditions.
The earliest realistic timeline for California nuclear power from new construction, assuming the moratorium falls in 2026, is approximately 2030–2033. That is meaningful progress against the FERC's 35 GW demand projection for 2030, but it requires the legislative and regulatory process to move without major delays.
More immediately relevant to AI power demand is whether lifting the moratorium changes where companies site data centers. If California becomes permissible territory for nuclear-adjacent siting — meaning data centers can be built knowing future nuclear supply is plausible — it may slow the migration of large AI workloads to states like Virginia and Texas that already have or are developing nuclear supply chains.
The legislation is a signal before it is a solution. But in energy policy, credible long-term signals change near-term investment decisions. California sending the signal that nuclear is back on the table matters even before a single reactor breaks ground.
Why did California ban nuclear power in the first place?
The moratorium was enacted in the late 1970s in response to the Three Mile Island accident in Pennsylvania and the broader anti-nuclear movement of the era. California's specific legislative framing conditioned any new nuclear construction on the federal government demonstrating a solution to permanent radioactive waste storage. Since that condition was never met, the moratorium functioned as an indefinite ban for 50 years.
What has changed to make California reconsider the ban now?
The primary driver is AI data center power demand. FERC projects U.S. data center electricity consumption will reach 35 GW by 2030, nearly double 2023 levels. California, home to the highest concentration of AI companies in the world, cannot meet that demand with renewable energy alone given intermittency constraints and available land. Nuclear is the only zero-carbon, always-on energy source available at the required scale.
What kind of nuclear technology would the new legislation allow?
The legislation would permit advanced nuclear technologies — primarily small modular reactors (SMRs) and other next-generation designs — that have received federal licensing since 2005. These are fundamentally different from 1970s-era reactor designs: smaller, factory-manufactured, using passive safety systems, and in some cases capable of using different fuel cycles that reduce long-lived waste.
How long would it take to actually build nuclear plants in California?
Even under an optimistic scenario, the timeline from legislative passage to operating nuclear capacity is 5–8 years. SMR designs have shorter construction schedules than conventional large reactors — 3–5 years per unit under ideal conditions — but permitting, environmental review, and site preparation add time. California's regulatory infrastructure has no recent experience processing nuclear applications, which will extend timelines further.
Are any major AI companies specifically pushing for California nuclear?
No AI company has publicly lobbied for this specific legislation, but the broader context is clear: Meta, Microsoft, Google, and Amazon have all signed major nuclear power deals in other states, demonstrating the industry's conviction that nuclear is necessary for AI power. Google's presence in California with major SMR commitments nationally creates an obvious alignment of interest with California's legislative effort.
What do environmental groups say about lifting the moratorium?
Environmental opposition remains significant but is divided. Traditional anti-nuclear groups oppose the reversal on grounds of accident risk, waste storage, and cost. But some environmentalists now support advanced nuclear as a climate necessity, arguing that the alternative — gas-powered data centers — is worse for the climate than nuclear. The debate within the environmental movement is genuinely contested in a way it has not been for decades.
Public opposition to AI data centers is growing fast. Energy costs, noise, water usage are driving community resistance.
Aikido plans submerged 100-kilowatt demo data centers off Norway's coast, embedded in floating wind turbines, to solve AI's energy and land-use crises.
A UK trial shows AI data centers can operate without continuous peak power, challenging hyperscaler cost models and opening new paths for energy-efficient AI infrastructure.